EMBER - Analysis of Malware Dataset Using Convolutional Neural Networks

Abstract

The aim of this research is to implement Neural Network algorithms to achieve a model of precision (f1-score and recall) for investigating malevolent Windows portable execution files. The paper utilizes EMBER - a benchmark dataset that contains features extracted from 1.1M binary files. The dataset contains 900K training samples (malicious, benign and unlabeled samples) and 200K test samples and provides numerous cases to build models that enhance information security. So, in order to determine if a given file is a malware or not we implemented algorithms like Convolutional Neural Networks and Feed Forward Neural Networks and assembled the results in terms of accuracy.